Automatic Classification of Cancerous Tissue in Laserendomicroscopy Images of the Oral Cavity using Deep Learning (1703.01622v2)

Abstract: Oral Squamous Cell Carcinoma (OSCC) is a common type of cancer of the oral epithelium. Despite their high impact on mortality, sufficient screening methods for early diagnosis of OSCC often lack accuracy and thus OSCCs are mostly diagnosed at a late stage. Early detection and accurate outline estimation of OSCCs would lead to a better curative outcome and an reduction in recurrence rates after surgical treatment. Confocal Laser Endomicroscopy (CLE) records sub-surface micro-anatomical images for in vivo cell structure analysis. Recent CLE studies showed great prospects for a reliable, real-time ultrastructural imaging of OSCC in situ. We present and evaluate a novel automatic approach for a highly accurate OSCC diagnosis using deep learning technologies on CLE images. The method is compared against textural feature-based machine learning approaches that represent the current state of the art. For this work, CLE image sequences (7894 images) from patients diagnosed with OSCC were obtained from 4 specific locations in the oral cavity, including the OSCC lesion. The present approach is found to outperform the state of the art in CLE image recognition with an area under the curve (AUC) of 0.96 and a mean accuracy of 88.3% (sensitivity 86.6%, specificity 90%).

• The paper presents a deep learning approach using CNNs to automatically classify cancerous tissue in oral cavity laserendomicroscopy images.
• Achieving an AUC of 0.96 and 88.3% accuracy, the deep learning model significantly outperforms traditional feature-based methods.
• This automated classification system has significant implications for improving early-stage oral cancer detection and aiding real-time clinical diagnosis.

Deep Learning for Automatic Classification of Oral Cancerous Tissue in Laserendomicroscopy Images

The paper presents a comprehensive approach to enhancing the classification accuracy of Oral Squamous Cell Carcinoma (OSCC) using deep learning applied to Confocal Laser Endomicroscopy (CLE) images of the oral cavity. OSCC, a prevalent cancer type affecting the oral epithelium, suffers from late-stage diagnosis due to inadequate early detection tools. CLE offers a promising imaging technique by capturing in vivo sub-surface micro-anatomical structures with high magnification and depth penetration, which is crucial for identifying malignancies approximately 100 microns below the tissue surface.

The researchers developed an automatic classification model leveraging deep learning, specifically Convolutional Neural Networks (CNNs), which significantly outperforms conventional textural feature-based machine learning techniques for this task. The paper utilized a data set comprising 7894 high-quality CLE images from OSCC patients and achieved outstanding discrimination performance, with a reported Area Under Curve (AUC) of 0.96 and a mean accuracy of 88.3%, coupled with a sensitivity of 86.6% and specificity of 90%.

Methodology Overview

• Data Acquisition: The CLE imaging dataset consists of 116 video sequences from 12 patients, acquired from diverse oral cavity locations. These images were preprocessed to exclude those impaired by artifacts or poor quality.
• Patch-Extraction and Data Augmentation: The images were split into patches to enhance feature extraction efficacy and reduce computational complexity. Data augmentation was employed to generate variations, such as rotations, to enrich the training dataset.
• Deep Learning Approaches:
  • Patch Probability Fusion Method: Utilizing CNNs, the method fuses patch-level classifications to produce an image-level diagnosis, significantly improving accuracy over traditional feature-based methods.
  • Transfer Learning: An Inception v3 CNN model, pretrained on ImageNet, was fine-tuned for CLE images to leverage existing knowledge from other image domains.

The experimental evaluation, employing a leave-one-patient-out cross-validation, further validated the model's robustness and accuracy. Comparisons against classical methods affirmed the superiority of deep learning techniques over those utilizing local binary patterns and gray-level co-occurrence matrices.

Implications and Future Directions

The presented methodology has significant implications for the diagnosis and treatment planning of OSCC:

• Clinical Utility: The development of an automated, rater-independent diagnostic system can enhance early-stage cancer detection, significantly improving treatment outcomes and potentially reducing cancer recurrence rates post-treatment.
• Real-time Applicability: The potential to integrate this system into surgical settings, offering real-time computer-aided diagnoses, can aid in accurately determining resection margins, thereby decreasing mortality and morbidity by reducing recurrence risks.
• Generalizability: While the focus is on OSCC, the approach may be generalized to detect other squamous cell carcinoma types in the upper aero-digestive tract, a direction worth pursuing in future studies.

Despite promising results, challenges in artifact identification, image preprocessing, and the need for larger histopathologically verified datasets remain. Future research could expand the model's capability to differentiate between dysplastic stages for early intervention.

In conclusion, this paper illustrates the profound potential of deep learning in medical image analysis, particularly in enhancing the accuracy and efficiency of cancer detection through advanced imaging technologies. The techniques outlined here contribute a significant step towards the automated, reliable, and scalable application of artificial intelligence in clinical oncology diagnostics.